I had to include more runs in N= 1024 with xm <= 0.64 and N:512 with 0.65<xm<=0.84 because the errors only start getting worse at around xm>= 0.5 and monomer number has much weaker effect which was harder to model without more larger aggregates data like N=1024.